Comprehensive modeling of cell culture profile using Raman spectroscopy and machine learning

Area of Science:

• Biotechnology
• Process Analytical Technology (PAT)
• Machine Learning in Bioprocessing

Background:

• Chinese hamster ovary (CHO) cells are critical for producing antibody-based therapeutics.
• Comprehensive metabolite monitoring is essential for optimizing CHO cell culture and ensuring product quality.
• Current Raman spectroscopy methods for CHO cell culture require time-consuming model construction, hindering rapid process development.

Purpose of the Study:

• To develop a simple, automated method for constructing accurate Raman spectroscopy models for CHO cell culture.
• To accelerate the analysis of numerous metabolites using machine learning and Python.
• To enable efficient process optimization and monitoring in pharmaceutical manufacturing.

Main Methods:

• Developed an automated workflow using Python and machine learning for Raman model construction.
• Employed Bayes optimization to automate and accelerate data preprocessing and spectral-range selection.
• Utilized various machine learning algorithms (linear regression, ridge regression, XGBoost, neural network) for model building, comparing them to partial least square (PLS) regression.

Main Results:

• Achieved improved model accuracy compared to traditional PLS regression.
• Demonstrated an automated approach for constructing Raman models for over 100 components.
• Successfully facilitated continuous monitoring of various parameters in CHO cell cultures.

Conclusions:

• The automated machine learning method significantly enhances the efficiency and accuracy of Raman spectroscopy model development.
• This approach enables comprehensive, real-time monitoring of CHO cell cultures, supporting faster process optimization and robust manufacturing.
• The study provides a valuable tool for the biopharmaceutical industry to improve antibody production processes.